Introducing Deeplinq for AML

Governance and AI aren’t mutually exclusive.

Start with a governed, billable AI engine. Extend the same controls across specialized applications, sovereign deployments, and future edge systems.

Request a demoExplore the platform

Built for regulated industries

The three pillars of the platform

Unified

One layer between your apps, data, and models. Keep standard OpenAI and Anthropic interfaces while providers change behind them.

Auditable

Security, policy, model, billing, and agent events enter a chained, exportable record designed for independent verification.

Governed

Identity, live grants, budget, configured guardrails, and tool approvals are enforced where each decision occurs.

What the engine unlocks

One platform. Four business outcomes.

The gateway, knowledge, agents, billing, and audit layers work as one system, so control becomes part of the product rather than a parallel project.

01

Create new revenue

Meter real usage, set model pricing and margin, fund customer credits, and stop calls before an empty balance incurs provider cost.

Billing ledger · runtime pricing · credit holds

02

Ship sooner

Connect standard model clients, configure governed knowledge, and run durable assistants without assembling a separate agent platform.

Gateway · knowledge · agents

03

Earn trust

Return cited answers, keep personal connections owner-only, and pause before a sensitive tool action executes.

Citations · consent · human approval

04

Show the evidence

Enforce tenant boundaries and export a tamper-evident record of identity, policy, model, billing, and agent activity.

Tenant isolation · audit integrity · recovery

Every action, on the record.

Watch agents and people work side by side. Every call is checked against your policies, attributed to whoever made it, and written to an audit trail you can replay.

An illustration of the Deeplinq console. It lists recent activity from agents and people side by side, showing for each call who or what made it, which policy was applied, whether it was allowed, held for approval, or blocked, and the audit record it produced.

Every token, metered.

Spend is tracked across every model you run, hosted or your own, and broken down by team, agent, and customer. Set hard caps so a runaway agent never runs away with the budget.

Spend by model

$5,320
this month
  • GPT-5.1OpenAI$2,140
  • Claude SonnetAnthropic$1,580
  • Mistral LargeMistral$690
  • Kimi K2Moonshot$420
  • Llama 3self-hosted$310
  • Gemini 3Google$180

One meter across providers. Switch or mix models without losing the bill.

Cost by team

Sample data
Reference month · % of cap
TeamTokensSpendCap
Fraud Ops8.4M$1,24062%
Customer Support24.1M$98048%
KYC / AML5.2M$1,61091%
Research3.1M$72034%
Tech Team12.6M$54071%
Batch summarizer31.0M$1,020capped

Where Deeplinq fits

Reference workflows

Reference patterns for regulated sectors, each grounded in controls the engine enforces today.

Banking · AML

Screen transaction alerts through governed models, with every decision policy-checked and attributed in the audit trail.

Explore workflow

Telecommunications

Govern support assistants across markets and settle model spend from measured token usage.

Explore workflow

Education

Ground student and staff assistants in approved institutional knowledge, with cited answers and access scoped to each user.

Explore workflow

Legal services

Give legal teams cited answers across approved matter files while client boundaries and sensitive actions stay under human control.

Explore workflow

Public sector

Serve citizens with AI that checks identity and policy before every action, and logs the verdict either way.

Explore workflow

Accounting

Review policies, spreadsheets, and supporting evidence through governed workflows where usage and approvals stay attributable.

Explore workflow

Where the core can go

Start with control. Expand without replacing it.

The product already unifies models, knowledge, agents, billing, and evidence. Its next markets add partner infrastructure, local distribution, specialized intelligence, and eventually connected operations around that same core.

Explore the company vision
Expansion pathNow · Next · Later

Now

Sell the control plane.

SaaS · pilots · managed deployments

Managed AI, secure knowledge, governed agents, controlled consumption, and audit evidence create the first recurring commercial offer.

Next

Deepen the application.

Industry AI · partner distribution · sovereignty

Domain partners turn the same core into specialized products while infrastructure partners widen regional delivery.

Later

Move into operations.

Edge · industrial AI · physical systems

The control model extends beyond software assistants into edge systems and physical agents where actions carry real-world consequences.

The architecture

Three layers, one governed gateway

Your apps on top, the Deeplinq engine in the middle, your systems and data underneath. Select a layer to look inside.

Screen the full path with a policy model.

A configured guardrail connector can inspect prompts, responses, retrieved evidence, and tool arguments for personal data, prompt injection, and unsafe content. Deploy the policy model where your boundary requires it.

Promptor agent action
Guardrail
Policy model

PII · Prompt injection · Safety

Any modelhosted or self-hosted
Guardrail
Policy model

PII redaction · Safety

Responsescreened, safe
Audit trail
18,432events

In your perimeter

Point the guardrail connector at a model inside the deployment boundary when policy requires local screening.

Fail closed when active

Once a guardrail connector is registered, a failing screening dependency returns an error instead of silently bypassing the control.

Logged either way

Allowed or blocked, every check lands in the append-only audit trail, attributed and timestamped.

FAQ

Answers for the security, deployment, and governance teams who have to sign off.

How secure is Deeplinq?
The engine enforces tenant scope with forced PostgreSQL row-level security, role-separated database access, organization-scoped vector shards, default-deny grants, and a cryptographically chained audit log. Credentials are sealed at rest. Edge TLS, infrastructure encryption, key custody, residency, and certification remain explicit deployment or organizational responsibilities.
Which models and providers can we use?
Deeplinq exposes OpenAI and Anthropic compatible interfaces and routes through registered connectors to hosted providers, compatible endpoints, or self-hosted models. Provider discovery helps register available models, with manual model IDs when a provider does not expose discovery.
Where does our data live?
The deployment chooses the database, object storage, vector store, region, and model endpoints. Self-hosting can keep the engine and local models inside a selected environment. Calls to hosted model providers create egress to those configured providers, so routing and deployment architecture must match your residency policy.
How does governance actually work?
The engine binds each request to a caller and immutable organization, checks live model and data grants, reserves budget, and applies configured guardrails. Agent tools are marked automatic or approval-required, and an approval-required action cannot run before its owner decides.
What does the audit trail capture?
The record covers authentication failures, access denials, grants, connector and pricing changes, billing, model routing, guardrails, knowledge decisions, agent lifecycle, and minimized external-call outcomes. Prompts, answers, raw evidence, secrets, and tool payloads are excluded.
How does Deeplinq control AI costs?
Usage is priced from measured input and output tokens. Before a call, the engine reserves its maximum cost against the organization credit balance, then settles actual usage and releases the remainder. Agent runs add explicit turn, time, and credit limits.
Do you train on our data?
The Deeplinq engine does not train a model on customer content. Retention and training terms for hosted models depend on the provider and account configured by the deployment. Self-hosted models avoid that provider egress path.
How do we deploy and integrate?
Point an OpenAI or Anthropic compatible client at the Deeplinq base URL and send the tenant credential issued for your application. The engine runs as one server with its admin console embedded, while durable stores and optional sidecars are configured for the deployment.